Every request to a shopify storefront crosses the edge first. It has to be fast, resilient and safe to change.
Work
Routing on Kubernetes, a GCP edge built on multi-tier load balancers and Rust proxy-wasm extensions, Cloudflare as a unified edge, and isolated canary rollouts for tier 1 services.
Outcome
Lower latency, cheaper compute and much faster incident mitigation. The numbers are on my resume.
cortex private, going open source
2026 · typescript, sqlite, unix sockets
A single-binary task tracker and memory layer for one engineer supervising several AI coding agents at once.
Problem
Parallel agent sessions lose context, attribute work badly and interrupt you constantly.
Decisions
One daemon per database on a private Unix socket, so every write is arbitrated and nothing is exposed to the network. Agents post updates through an agent-first CLI with --json. Live events fan out to every open terminal workspace. The TUI only interrupts when a task is blocked. Tasks double as searchable agent memory.
Outcome
Daily driver for my own agent work: about 68k lines of TypeScript with a roughly 1:1 test-to-source ratio (1,600+ tests).
Two open-source extensions for the pi coding agent. pi-minions delegates work to isolated sub-agents you can watch and halt. pi-quests keeps long sessions on the original goal with a quest log and rollback checkpoints. Both have outside users and issues from people I have never met.
A cross-platform dotfiles installer with a curl bootstrap and benchmarked shell startup, maintained for five years. A documented 15U home rack with VLAN-segmented networking, DNS sinkhole and Home Assistant.
Experience
staff infrastructure engineer @ shopify
2025 – present
edge infrastructure
senior infrastructure engineer @ shopify
2021 – 2025
edge infrastructure
data replication software developer @ IBM Canada
2019 – 2021
change data capture for kafka, and automation for the IBM cloud offering.
